Most job descriptions narrow the pool by accident. A mid-level role grows a ten-year requirement, a wishlist of "nice-to-haves" reads as hard requirements, and gendered words like "rockstar" or "aggressive" quietly tell whole groups not to apply. This system takes your rough notes and produces a JD that screens for the skills the job actually needs — and flags every phrase that might be turning good people away for no real reason.
- Your rough notes on the role: title, level, team, and what the person will actually do day to day — a few bullets typed straight into the chat in Claude Desktop, or dropped in as
role-notes.md, is plenty. - The genuine must-have skills (the 3–5 things someone truly can't do the job without) kept separate from the nice-to-haves.
- A JD you've used before — even an imperfect one, dropped into the chat as
last-jd.md— so Claude can match your format and catch your house habits.
-
Separate the real must-haves from the wishlist
Before any drafting, force the requirements list to shrink. Every extra "required" line narrows the pool, so the job here is to demote anything that isn't genuinely load-bearing.
you askRead role-notes.md. This is a mid-level role. Split what I wrote into two lists: genuine must-have skills (someone literally cannot do the job without these) and nice-to-haves. Challenge anything that looks like a wishlist item dressed up as a requirement, and flag any years-of-experience number that seems too high for the level. Don't write the JD yet.what you get back Two clean lists — maybe 4 true must-haves and 7 nice-to-haves — plus a note like "'8+ years' is high for a mid-level role; consider 3–4" so you can decide before it's baked into a posting.
This split is the highest-leverage move in the whole system: a shorter must-have list is the single biggest thing that widens the pool.
-
Draft the description from the trimmed list
Now generate the JD, but anchor it to the must-haves you just confirmed so the requirements section stays honest instead of creeping back to a wishlist.
you askNow write the full job description using only the must-haves we confirmed, with nice-to-haves clearly labeled as bonuses. Match the structure of last-jd.md. Sections: what the role does, what you'll work on, must-have skills, bonus skills, how we work. Plain English, no buzzwords.what you get back A complete JD with the requirements section visibly shorter than usual — must-haves and a clearly-optional "bonus" block — in your existing format and tone.
-
Scan for language that screens people out
This is the inclusion pass. Have Claude hunt for the wording that quietly narrows your pool — and explain why each flag matters, so you're learning the pattern, not just patching one posting.
you askNow scan this JD for anything that might narrow the applicant pool for no real reason: gendered or coded words (rockstar, ninja, aggressive, he/she), an experience bar that's too high, jargon a great candidate from a different background wouldn't recognize, and any 'culture fit' phrasing. For each flag, quote the phrase, say why it could screen people out, and suggest a neutral rewrite.what you get back A table of flags — "'rockstar developer' → reads as masculine-coded, try 'strong engineer'", "'must thrive in a fast-paced environment' → vague, say what that actually means" — each with a quoted phrase, the reason, and a fix you can accept or reject.
Read every flag yourself. Some "coded" words genuinely fit your role; you decide which to keep — Claude surfaces, you choose.
-
Add an inclusive close and a clear apply path
The last lines of a JD change who hits apply. A note that encourages people who don't tick every box, plus a concrete next step, recovers candidates the requirements list scared off.
you askAdd a short, genuine encouragement-to-apply line for people who don't meet every bullet (no clichés), a one-sentence note on how we support flexibility or accommodations, and a clear 'how to apply' with what to send and a [closing date] placeholder. Keep it warm and specific.what you get back A closing block that explicitly invites people who meet most-but-not-all requirements, names your flexibility honestly, and ends with a concrete apply step and a date placeholder you fill in.
- Senior or specialist role: keep the must-have list short but make it sharper — ask Claude to phrase each requirement as a demonstrable skill ("has shipped X") rather than a years number, which screens for ability instead of tenure.
- Anchor to the competency framework: if you've built one, pull the must-haves straight from your Build the competency framework every role inherits doc rather than reinventing them per posting — the JD inherits the bar, so what you post is what you'll later score and promote against.
- Arabic & bilingual postings: for an Arabic-language role, write the JD natively in Arabic and run the same bias scan in Arabic — adapt, don't machine-translate, because coded and exclusionary language is specific to each language and an English-then-translate pass sails right past it. Have a fluent person own the final wording.
- Reuse it across the team (Power Track): once your prompts work, save them as a
/jdcustom command (see the Playbook's Capabilities tab) so every hiring manager gets the same inclusion scan, not just the careful ones. Custom commands are the opt-in Power Track — on Desktop you can share the dialed-in prompt as a saved note any manager pastes in. - Feeds the hiring system: this JD is step one of the Run a hire from open to offer capstone — the trimmed must-haves become the spine of the interview scorecard, so the questions you ask map to the skills you posted.
- A bias scan is a prompt, not a guarantee. Claude flags common patterns; it won't catch everything, and it can over-flag a word that genuinely fits your role. Read every flag and make the call yourself.
- Don't let nice-to-haves quietly drift back into the must-have list between drafts — re-check the requirements section before posting, because that creep is exactly what narrows the pool.
- Claude drafts the JD; a human owns what the role actually requires and what's legal to ask for. Anything touching protected characteristics or local hiring law goes past a real person before it's published.
you'll end up with A clear, inclusive job description in about twenty minutes — built from genuine must-haves, scanned for the wording that quietly screens people out, and ready to open a role without narrowing the pool by accident.
Questions people ask
- What do I need to prepare before starting?
- A few bullets in `role-notes.md` covering the title, level, team, and what the person will actually do day to day. You also need a previous JD in `last-jd.md` so Claude can match your format. That's it — the notes don't need to be polished.
- Does the bias scan catch everything?
- No. Claude flags common patterns — gendered words, high experience bars, vague culture-fit phrases — but it won't catch every form of exclusionary language, and it may flag a word that genuinely fits your role. Read every flag yourself and make the call; the scan surfaces, it doesn't decide.
- Can I adapt this for a senior or specialist role?
- Yes — keep the must-have list short but make each item sharper by asking Claude to phrase requirements as demonstrable skills ("has shipped X") rather than years-of-experience numbers. This screens for actual ability rather than tenure, which is especially useful at senior levels.
- What happens if nice-to-haves drift back into the must-have list between drafts?
- That creep is exactly what narrows the pool. Before posting, re-read the requirements section and compare it against the confirmed must-have list from step 1 — anything that wasn't on that list shouldn't be in the required section.